Ask just about anyone what makes living with diabetes so challenging, and they think they know the answers.
“The shots.” (Actually, you get used to them pretty quickly.)
“The meal planning.” (Truthfully, it’s what everyone should be doing.)
“The cost.” (Yes, that’s pretty rough, but not the end-all.)
The reality? What weighs most on those with diabetes or battling pre-diabetes is more than just a treatment plan, countless needle pricks, or even the oft-daunting bills.
“It’s the endless, constant, and inescapable buzzing in your head about what you’re doing right, what you’re doing wrong, how long your insulin has been on board, if the barista really measured the peanut butter like you asked,” says Jaime Jones of Colorado, who has been managing her child’s type 1 diabetes (T1D) for eight years now. “It’s the inescapability of it all. Honestly, it can feel soul-crushing.”
Indeed, experts say people with diabetes are in constant decision-making mode. Famed endocrinologist Dr. Howard Wolpert, Chief Medical Officer for Connected Care at the Lilly Cambridge Innovation Center, estimates a person with diabetes makes at least 300 diabetes-related decisions a day. That’s more than 109,000 decisions per year.
No wonder folks get burned out, tripped up, or just plain quit.
Could all that be moving toward a better tomorrow? The use of so-called machine learning, or artificial intelligence (AI), in diabetes care is a growing industry. It’s focused not just on better outcomes for those with diabetes, but a better life as they work toward those goals.
The hope is to create accessible programs, apps, and other tools to take over some of the constant decision-making — or, at the least, help people gather and assess their own data to make sense of it in a way that helps ease their day-to-day burden.
Could AI be the next big breakthrough in daily diabetes care? Many are hoping so.
The terms AI and machine learning are often used interchangeably because they refer to the capability of computers to crunch enormous data sets and “learn” from the patterns detected at a level that the human mind could never achieve.
Artificial intelligence in healthcare is expected to be an $8 billion industry by 2022, but the Food and Drug Administration (FDA) still has concerns about the reproducibility of results and flaws in the datasets used — including lack of diversity.
Still, the dream is an AI-powered world where the step tracker talks to the menstrual cycle calendar, the heart monitor, the meter or continuous glucose monitor (CGM), and more. These systems would share and compare data using algorithms and then present, in an easy-to-read-and-access, simple-to-understand way, what decision would be best for a person in the moment, like a doctor in your pocket or a true “diabetes whisperer,” quietly guiding you toward those decisions and freeing your brain to focus on the rest of your life.
Pipe dream? Perhaps not.
In the diabetes world, AI has already enabled a revolution in closed-loop systems (aka Artificial Pancreas technology) and interconnected tools to help a person with diabetes gather and store more data, see trends from that data, and be pointed to better decisions.
When insulin pumps first started keeping track of things like bolus doses for past meals, the diabetes world celebrated. It was a baby step, and now additional steps have brought us to smarter and more integrated tools.
Today, companies like Livongo, Cecelia Health, One Drop, Virta Health, and mySugr are all up and running with AI-powered systems designed to help gather, store, disseminate, and utilize data for more efficient and individualized diabetes care.
Livongo, for example, combines blood sugar monitoring with coaching and remote monitoring (nudging the user when need be), along with some nice touches like keeping track of how many strips you use and reminding you to order. One Drop helps users track glucose levels along with activity, medications, and food, offers in-app coaching, and connects users to a community for support when needed. Virta Health offers virtual nutrition coaching for those with pre-diabetes and type 2 diabetes.
The fun tagline at mySugr embodies the goal of them all: “To make diabetes suck less.”
Their system comes in three levels. First, a free app that guides users in tracking glucose levels, insulin doses, meals, and more, and then offers a detailed analysis of that information. It estimates A1C results, prints out a report for medical appointments, and gives users a solid look at 24 hours of information at any time.
There’s also a higher-level, more amped-up report, and a third level of service that brings in coaching in the form of diabetes educators who watch and study users’ information and reach out when they deem it needed.
Longtime type 1 Scott Johnson, who is mySugr’s spokesman, says he would not call it “true AI” yet, but said the company is on track to get there in time.
“We know diabetes care is data-driven,” Johnson says. “But really, not many people continue logging (data) for long. mySugr does that kind of work now. And in the future, it will offer even more data analysis and guidance.”
He adds on a personal note, “I want to offload as many of my diabetes decisions as possible, and frankly, I think [mySugr] can do a better job of it than I can.”
There’s widespread consensus that while these are better than anything previously available, AI could go much further towards improving life with diabetes.
London-based Cyndi Williams was working as a chemical engineer and innovator when she met co-worker Isabella Degen, who happens to have T1D herself. In time, the two realized they had a combined calling: Create a platform that betters the lives of those with diabetes and those who care for them.
Quin stands for “quantifying intuition,” which is a nod to anyone using insulin. While the developers don’t plan for this to be a closed loop technology, it does include many of those automated and decision support functions that APs can offer.
What Quin does — or what Williams and team are working for it to do — is take every kind of personal health data possible, morph it with daily life decisions, and then use all that combined information to help people with diabetes make smart choices with less brainwork.
In time, Williams says the app will dig deep into the many physiological and psychological happenings in a person’s body, track what different foods do to a person at different times and in different places, take all that as one, and become, in essence, that all-knowing doc in a pocket people with diabetes may need.
While not yet available in the United States, an early version has been in the hands of users in Ireland and the UK for the past year.
Importantly, Quin does not demand a person be on an insulin pump or even a CGM. It does not study or suggest carb ratios, nor does it predict blood sugar levels.
“Until now, digital diabetes has been very much about observing what we do and putting it in the data. It’s relatively flat,” Williams says. “We live in a world where Spotify knows what music we want to listen to. We are not yet there in diabetes, but we can be. We want to reduce the cognitive load on a person with diabetes.”
Quin draws on data from other health tools a person with diabetes may use (step trackers, heart rate monitors, etc.) and also from the information they share directly with the app to help formulate decisions based on past life experiences.
In other words, Quin helps the user decide what to do in the moment, based on intelligence gathered from past similar decisions. It does all the work for you: Instead of scouring your brain for the “What the heck happened that other time I had a latte at high noon?” you can look to Quin to do that memory work, overlay it on the current situation, and seamlessly zero in on an action decision.
Their algorithm is dependent on some input: Quin asks the user to take a photo of a meal (or that latte) and enter that info. Quin will go from there and mark other data points: the time of day, your heart rate, if you are busy or stressed, and more. Then it helps you see not just what amount of insulin dose may be best for that food, but what dose is best for that food at that moment for you and only you.
“It’s a philosophy based on the idea that your past decisions (no matter their outcome) are the best information we have,” Williams says.
Although things like lower A1Cs and more time in range (TIR) are crucial, the goal goes beyond blood sugars, she says. “What we are looking at is how can we improve the entire life of the person.”
User results have been strong so far. A pre-clinical trial in spring 2019 including 100 users showed that 76 percent had fewer hypos and 67 percent had better TIR. Also, over 60 percent said they “feel more confident and report that their life with diabetes is better now,” Williams notes.
They will likely go through the process to apply for insurance reimbursement and hope to make the app available in the U.S. by 2022.
“We see this as a long journey,” she says. “We see Quin becoming smarter and smarter and doing that cognitive physiological offload. We see it bringing better emotional health.”
Biotech and business expert Noosheen Hashemi was attending a medical conference at Stanford University shortly after a conference on machine learning when she had the idea for January.ai, a new AI-based support system designed to empower people with type 2 and pre-diabetes. In particular, she was inspired by the patients who had shared their stories at the Stanford conference.
“They resonated with me. What they said was this: ‘Look at the whole person instead of reducing people to a single marker,'” she says.
That’s the foundational goal of January.ai: AI to help each individual adapt their lives and treat their diabetes in their own unique way. The platform will combine data from different wearables along with information input by users on their own biology, needs, and even, yes, desires.
Hashem explained that everyone differs so much in their glucose responses to food, even in ourselves between varying situations. That “impossible hurdle” of navigating a food response is what January.ai is tackling.
“Not everyone can drop 25 pounds when asked to,” she says, but with the right focus, information, and guidance, “Everyone can manage their blood sugars.”
When the platform launches ideally some time this fall, new users will be able to signup for a four-week program called the “Season of Me” that will include helping them obtain a CGM to track glucose trends. Hashemi says they have a network of providers in place who can help with prescriptions — even though their initial focus is not insulin users but pre-diabetes.
For the first two weeks, the combined CGM and platform features will help users learn how their own body and blood sugars react to certain foods and activities. In the following two weeks, their system guides users on how to integrate that learning into a daily routine.
January.ai is a true learning platform, so the longer you use it, the more helpful it is. For example, if you want to go out to eat and know which burger you plan to order at a particular restaurant, the system can search your history to see if you’ve had it before, along with what other things were going on in your body and life at that moment, and how your blood sugar responded.
Each meal and instance helps January.ai learn more, and thus be ready to help even more over time.
The system also presents healthy alternative options: What if you skip the bun? (It shows you a probable outcome). Is there another menu option that’s similar but perhaps with fewer carbs or calories? It even offers users ways to “earn” a treat or the occasional splurge food that those in the T1D community often call “bolus-worthy.”
For instance, it might suggest you go for the burger with the bun and then, based on what it knows about you, suggest a timed walk just after.
“We are hyper-focused on the user experience,” Hashemi says. “Let’s first help some people. And if we can delight them somehow, give them new insights on how to savor life while making smart choices, we’re winning.”
Admittedly, Quin and January.ai sound pretty sci-fi. Can this technology really function to change people’s daily experiences?
For early adopters, it might not be a stretch. But even for those not tech-savvy, developers believe the time is ripe.
One of those is LaurieAnn Scher, a diabetes care and education specialist (DCES) who serves as Chief Clinical Strategy Officer at Fitscript, a digital health company providing online fitness programs for diabetes and other chronic conditions.
“Tech is something that, as diabetes care specialists, can help us take a great leap,” she says. “Sometimes the right person just hasn’t been exposed to it yet.”
Scher points out that, at best, people grappling with diabetes generally see a healthcare provider just four times a year, and it’s not like diabetes needs ebb between those times.
“These apps have a great way to fill the gaps and help stop things if something is brewing,” she says. “I wish I could be… available to patients 365 days a year, 24 hours a day. But I cannot be. This will fill the gaps when providers are not available.”
Another advantage is that dealing with data and facts, AI-based tools take the emotional bias out of diabetes management. Rather than facing some medical school-educated professionals who may seem to be judging you, you’re just looking at the data and recommendations on a neutral basis.
Scher admits that, sometimes, using an app or platform can feel burdensome. But AI provides long-term advantages: As the system learns more about you, it can help you more and remove the burden.
“It’s more work, but it’s useful work,” she says.
Chris Bergstrom, a former BD and Roche Diabetes Care executive and former Head of Digital Therapy at Boston Consulting Group, sees good in the AI future.
“Today, treatment algorithms are mostly one-size-fits-all based on… thousands of patients. Tomorrow, through digital health, these algorithms will be based on millions of people in the real world. Using AI, we can then enable a level of personalization otherwise unimaginable,” he says.
“Which drug, which device, which diet is right for ME based on my genetics, co-morbidities, lifestyle, motivations, economic resources, and other social determinants? (AI) unlocks the power of population data to drive personalized diabetes care,” continues Bergstrom. “It will be a game-changer.”
In other words, perhaps the collective brains of millions of people with diabetes will have freed space when they no longer need to calculate for every meal and activity. Who knows what could come from that?